Berlekamps Switching Game on Finite Projective and Affine Planes

I adapt Berlekamp’s light bulb switching game to finite projective plans and finite affine planes, then find the worst arrangement of lit bulbs for planes of even and odd orders. The results are then extended from the planes to spaces of higher dimension.

💡 Research Summary

The paper revisits Berlekamp’s classic light‑bulb switching puzzle and transports it from the familiar rectangular grid to the realm of finite geometry, specifically finite projective planes PG(2,q) and affine planes AG(2,q). In the original game a player may toggle any row or column, which flips the state of every bulb lying at the intersection of that row and column. The objective is to turn off as many bulbs as possible; the “worst” initial configuration is the one that maximizes the number of bulbs that remain lit after any sequence of switches.

The author first formalises the game on a projective plane of order q. A projective plane contains q²+q+1 points and the same number of lines; each point lies on q+1 lines and each line contains q+1 points. The natural analogue of a “row” is a line, and a “column” is simply another line – the game therefore has one switch per line. Pressing a switch toggles the state of every point on that line. By representing the bulb configuration as a vector over the finite field GF(q) and each switch as adding the incidence vector of a line, the whole system becomes a linear algebraic model: the incidence matrix I (size (q²+q+1)×(q²+q+1)) encodes the effect of all possible switches.

The central question is: for a given q, which initial vector v maximises the Hamming weight of any vector in the coset v+Im(I)? In other words, what is the configuration that cannot be reduced below a certain number of lit bulbs, regardless of the switch sequence?

When q is even, the incidence matrix has full rank over GF(2) (the parity field relevant to toggling). Consequently every non‑zero vector in the configuration space can be expressed as a sum of line‑vectors, but the all‑ones vector (all points lit) lies outside the image of I. No combination of line toggles can turn all bulbs off, and the all‑ones configuration is provably the worst case: any switch sequence leaves at least q²+q+1 bulbs lit. The proof uses the fact that each line contains an even number of points, so toggling any line changes the parity of the total number of lit bulbs by an even amount; starting from an odd total (all‑ones) the parity can never become zero.

When q is odd, the situation changes dramatically. Each line now contains an odd number of points, so toggling a line flips the overall parity. The author shows that the configuration with exactly one point dark and all others lit cannot be reduced further. The argument proceeds by examining the kernel of I over GF(2): the kernel has dimension q+1, corresponding to the set of line‑combinations that leave every point unchanged. Any attempt to eliminate the single dark point would require a combination that flips an odd number of points on each line through that point, which is impossible because those q+1 lines form a basis for the kernel. Hence the worst configuration for odd q consists of all points lit except one.

The paper then turns to affine planes AG(2,q), which can be obtained from PG(2,q) by deleting a line at infinity. In this setting there are q² points arranged in a q×q grid and 2q lines (q horizontal, q vertical). The same parity arguments apply. For even q the all‑lit grid is worst; for odd q the worst configuration is the grid with a single row and a single column left dark (equivalently, all points lit except those on a chosen “central” line of each direction). The author demonstrates that the removal of the line at infinity reduces the rank of the incidence matrix by one, but the parity‑based classification of worst cases remains unchanged.

Finally, the author generalises the results to higher‑dimensional finite projective spaces PG(n,q) and affine spaces AG(n,q). In PG(n,q) there are (q^{n+1}−1)/(q−1) points and the same number of hyperplanes; each hyperplane contains (q^{n}−1)/(q−1) points. The switch set is now the collection of all hyperplanes. Using the same linear‑algebraic framework, the author proves that for even q the configuration with every point lit is worst, while for odd q the worst configuration is “all points lit except those on a single hyperplane”. In the affine case the same pattern holds after removing a hyperplane at infinity.

The paper concludes by discussing implications for combinatorial design theory and coding theory. Projective‑plane based designs (e.g., symmetric BIBDs) and their associated error‑correcting codes often rely on the incidence structure of points and lines; understanding the extremal toggle configurations sheds light on the robustness of such designs under parity‑preserving transformations. Moreover, the toggle model provides a concrete visualisation of group actions of the projective linear group PGL(n+1,q) on point sets, suggesting further research directions such as randomised switch strategies, extensions to non‑Desarguesian planes, and connections with spectral graph theory.